[NOOK-176] OCR 장소 후보 추출 상한 확대

[dh1010a]

목적

OCR transcript에 다수의 장소명이 포함된 게시물에서 장소 clue 추출 단계의 개수 제한 때문에 후보가 과도하게 탈락하지 않도록 개선합니다.

변경 내용

• 장소 clue 반환 상한을 20개에서 60개로 확대
• 본문 기대 장소 수 인식 상한을 20개에서 80개로 확대해 52곳 같은 게시물을 OCR fallback/recovery 판단에 반영
• OpenAI structured output schema의 place maxItems를 60개로 확대
• OpenAI place clue 출력 토큰 한도 상향
  • text clue: 800 → 2500
  • image transcript clue: 4000 → 12000
  • content inference: 3000 → 8000
• 프롬프트의 “최대 20개 가게” 문구를 “최대 60개 가게”로 갱신
• 회귀 테스트 추가
  • 20개 초과 image place clue 처리
  • 52곳 본문에서 text result가 부족하면 OCR fallback 수행

검증

• ./gradlew :nook-api-application:test --tests org.every.nook.api.application.place.ProcessPlaceParsingJobUseCaseTest
• ./gradlew :nook-api-infrastructure:test --tests org.every.nook.api.infrastructure.openai.OpenAiContentInferenceAdapterTest
• ./gradlew clean check

참고

• develop에는 동일 변경을 0c6e492로 반영했고 push 완료했습니다.
• main PR 브랜치는 origin/main 기준으로 NOOK-176 변경만 cherry-pick했습니다.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 84d716fafd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Keep the processing lease valid for the enlarged workload

When a post now yields dozens of clues, resolveClues can perform up to four sequential provider searches plus candidate selection per clue, so processing can exceed the default one-minute lease. The presentation worker still dispatches outstanding jobs every 10 seconds, while PlaceParsingPersistenceAdapter.findOutstanding marks a PROCESSING job available solely from its last updatedAt; there is no heartbeat during resolution. A second worker can therefore reclaim the same high-count job while the first is still running, causing duplicate provider work and competing complete transactions that can hit post_places uniqueness constraints or overwrite the job status. Increase/renew the lease or otherwise prevent reclamation while processing the newly supported workload.

Useful? React with 👍 / 👎.